ietf-nntp New wording on article numbers - draft 3

[Alan Barrett]

> >Let me see if I understand this correctly.  Are you saying that
> >a collection of articles numbered, say, {12,13,14,15,16,17}
> >at the master, could arrive at the slave in, say, the order
> >{13,15,12,17,14,16} ? But all earlier articles would arrive before
> >this collection, and all later would arrive after -- so the out of
> >order arrival is localised ?
>
> um, I don't want to be the stick in the barrel, but are there any
> circumstances where the above behaviour would actually manifest itself
> in reality? if so, could you give a good example, complete with
> reasons at all of the points?

Consider a "master" server, which receives and sends ordinary news feeds
to one or more peers, just like an ordinary news server.

Consider a "slave" server which wishes to use the same article numbers
as the master server.  To this end, the slave receives a news feed from
the master, but does not receive a feed from any other news system.  The
slave learns the article numbers that the master has used (either by
looking at Xref headers, or via a protocol extension such as xreplic),
and the slave then uses the same article numbers as the master.

If the feed from master to slave consists of a single NNTP session, then
articles from the master to the slave will usually be transmitted in
order, and there is no problem.  But in practice articles do sometimes
get out of order.  For example, a temporary failure could result in
article 12 being transmitted successfully, article 13 failing and being
written to a backlog file to be tried again later, and article 14 being
transmitted successfully.

If the feed from master to slave includes several parallel connections,
then articles 12 and 13 might be transmitted over different connections,
and might arrive out of order, even in the absence of any temporary
failures.  Guaranteeing that all articles in a particular newsgroup are
transmitted over the same connection is infeasible in the face of
crossposted articles.

> (the hard point with doing pure theoretics is that we don't always
> think of applications right off the bat... sharing the applications
> can help make the theory a lot easier)

I hope this helps.

--apb (Alan Barrett)